This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [patch 04/12] entryval: Virtual tail call frames
- From: Daniel Jacobowitz <drow at false dot org>
- To: Jan Kratochvil <jan dot kratochvil at redhat dot com>
- Cc: gdb-patches at sourceware dot org
- Date: Mon, 18 Jul 2011 16:56:31 -0400
- Subject: Re: [patch 04/12] entryval: Virtual tail call frames
- References: <20110718201658.GE30496@host1.jankratochvil.net>
On Mon, Jul 18, 2011 at 4:16 PM, Jan Kratochvil
<jan.kratochvil@redhat.com> wrote:
> Between frame #3 and frame #4 is function `amb' but one cannot reliably
> reconstruct its exact PC. ?The code tries to determine only a sequence of
> unambiguous bottom calees and top callers and anything in between is dropped.
> Currently GDB knows that between frame #3 and frame #4 it is uncertain.
> Former patch tried to display some "???" there. ?It had a real frame_info,
> later I removed it, I think such "???" marker should never have its real
> frame_info inside GDB. ?Still the backtracing function could display some
> uncertainty indicator for the user; this is currently not implemented.
> GDB already behaves such way without trying to recover any frames info.
For the record, based on my experiments with inlined function
unwinding, I agree with your conclusion here. We shouldn't keep a
frame_info around unless it is a part of the default backtrace, and we
shouldn't display a frame that lacks accurate PC and source location
information.
I haven't looked at the patches, but I'm impressed with what you've
accomplished here!
--
Thanks,
Daniel